Hmm, perhaps I'm reading too much into this, but is this a side effect of the majority of the training data being from earlier years - much like I tend to vaguely think of "50 years ago" as being the 1950s if I'm not really paying attention...? :)
I believe so, see my result with Haiku extended thinking on. I think the weights are just too biased towards blurping out the majority of the training data of 'next year is xxx'. Interesting problem to solve indeed.
